# Stereo vision

two aligned cameras plus neural depth perception—to determine 3D depth and spatial structure. Proces depth through software algorithms rather than laser-based sensing.

# Stereo Vision Basics

[![image.png](https://bookstack.hku.nl/uploads/images/gallery/2025-09/scaled-1680-/6Bhimage.png)](https://bookstack.hku.nl/uploads/images/gallery/2025-09/6Bhimage.png)

# Zed 2i

{{@242}}

# OAK-D (Lite)

A small depth camera that combines **stereo vision** with an onboard **AI processor.**  
The **OAK-D** can be used for 3D scanning because it produces **RGB + depth maps** that you can turn into a point cloud and mesh. You move the camera around an object or space, capture overlapping frames, and then stitch them together with software (e.g. Open3D, MeshLab, or ROS).

But: it’s **not the preferred tool** for 3D scanning.

- Accuracy is lower than LiDAR or photogrammetry.
- Shiny/transparent surfaces don’t scan well.
- Range is limited (~10 m).

It’s great for **real-time depth perception and robotics**, but for **high-quality 3D models** you’d usually go with photogrammetry or LiDAR.  
  
\---------------------------------------------

- **How it works:**
    
    
    - Two synchronized monochrome cameras capture slightly different views of the same scene.
    - Depth is computed by comparing the disparity (shift) between the two images.
    - An onboard neural depth engine accelerates the calculations, so the host computer doesn’t have to.
    - Optionally, an RGB camera provides color overlays on the depth map.
- **Key features:**
    
    
    - Depth range: ~0.3 m – 10 m.
    - Field of view: ~70–80°.
    - USB-C powered, plug-and-play.
    - Runs AI models (object detection, face recognition, body pose estimation) **directly on the device**.
- **Why it matters for stereo vision:**
    
    
    - Demonstrates that stereo vision can be compact, affordable, and real-time.
    - Doesn’t need external GPUs/CPUs for heavy lifting.
    - Provides a good bridge between basic stereo rigs (like DIY dual webcams) and advanced research hardware (like ZED cameras).
- **Use cases:**
    
    
    - Robotics navigation.
    - Human pose tracking.
    - AR/VR prototyping.
    - Object recognition combined with depth.

Here's info on how to use it in Touchdesigner: [https://derivative.ca/UserGuide/OAK-D](https://derivative.ca/UserGuide/OAK-D)  
The example file is `OAKExamples<span class="mw-lingo-term" data-lingo-term-id="76746356f2d9fd021f01bc051bb2114f">.toe</span>`. Instructions and tips are inside the file. The file is located in `C:/Program Files/Derivative/TouchDesigner.2023.xxxxx/Samples/OAK`.